For that visualization, I filtered the dataset to only have one (random) painting per artist per year (some artists in the sample have dozens per year). Also, all PCs are centered at 0, but here showing absolute values (to highlight any deviations from the average, in both directions), and also capped at |20|, as there are some outliers in PC1 and PC2 (so most other would be all gray otherwise). I also tried normalizing the PCs, but then PC1 becomes mostly gray, since it’s easier to be near the “edges” in other PCs.


The PC distributions with examples for reference. The peaks are always at 0. Now updated with more representative sample paintings (also note that the colorfulness measure in here is only about n unique shades, hence the brown-shades-only dudes painting on the high end; but I implemented a proper colorfulness/range measure for the next run).